On an e-commerce platform, each user session can include multiple actions—viewing products, adding or removing items from the cart, and completing purchases. Estimating the monetary value of each session (session_value) is critical for:
- Understanding which sessions are most profitable
- Prioritizing high-value customers
- Detecting anomalies (e.g., unusually high or zero-value sessions)
- Optimizing business strategies like remarketing and personalized offers
However, predicting session value is challenging because:
- Many sessions include only product views without purchases
- Event sequences can be inconsistent (e.g., a purchase without adding items to the cart)
- User behavior varies widely—from one-time visitors to loyal repeat customers
The goal of this project is to predict the session value for each user session using behavioral and event-based features. Our approach includes:
- Conducting Exploratory Data Analysis (EDA) to understand user, session, and product-level activity
- Detecting anomalies such as unusual sessions, order inconsistencies, or suspicious behavior
- Feature engineering from session sequences to capture meaningful patterns
- Training machine learning models (CatBoost and AutoGluon) to predict session value
This notebook combines data exploration, anomaly detection, feature engineering, and machine learning modeling to generate actionable insights for session-level revenue prediction.